memsearch
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill indexes and processes external Markdown files, creating a vulnerability surface where malicious instructions embedded in the data could influence agent behavior.
- Ingestion points: Markdown files within the user's specified memory directory (e.g., in
SKILL.mdandreferences/upstream.md). - Boundary markers: Instructions mandate a clear separation between canonical Markdown data and the derived Milvus index, and require explicit approval before acting on indexed data.
- Capability inventory: The skill utilizes
Bashto execute CLI operations includingindex,search,compact, andreset. - Sanitization: The skill does not perform automated sanitization of the input data, relying instead on user confirmation for all data-processing steps.
- [COMMAND_EXECUTION]: The skill includes a security-focused script (
scripts/preflight.py) for verifying the integrity of the local environment. - The script implements secure POSIX directory traversal using
O_NOFOLLOWanddir_fdto mitigate symlink race conditions (TOCTOU). - It performs SHA-256 hash verification of selected files against a manifest in
references/source.json. - It contains an explicit blocklist for sensitive paths, including
.ssh,.aws,.env, and credential-related file stems. - [SAFE]: The skill follows best practices for external tool integration by pinning specific repository commits and providing a manifest for offline verification. All high-risk operations, such as network access, destructive resets, and plugin installations, require explicit user approval.
Audit Metadata